Means of a Dirichlet process and multiple hypergeometric functions. (Q1879833): Difference between revisions

Distributions of random means of measure-valued gamma amd Dirichlet processes are described. Given a finite measure \(\alpha\) on \textbf{R}, a measure-valued random element \(\gamma\) is called a gamma process with the parameter \(\alpha\) if for any collection of sets \(B_1\),\dots,\(B_k\) the random variables \(\gamma(B_1)\),\dots,\(\gamma(B_k)\) are independent and gamma-distributed with \(\mathbf E\gamma(B_i)=\text{Var}(\gamma(B_i))=\alpha(B_i)\). Analogousely \(\varphi\) is a Dirichlet process with the parameter \(\alpha\) if \((\varphi(B_1),\dots,\varphi(B_k))\) has a Dirichlet distribution with parameter \((\alpha(B_1),\dots,\alpha(B_k))\). Different characterizations of random means \(\varphi(I)=\int x\varphi(dx)\) and \(\gamma(I)=\int x\gamma(dx)\) are given. Lévy-Khinchin representation for the characteristic function of \(\varphi(I)\) is derived. It is shown that any finite \(\alpha\) with \(\int \log(1+|x|)\alpha(dx)\) is determined by the distribution of \(\varphi(I)\). The distribution of \(\varphi(I)\) is symmetric iff \(\alpha\) is symmetric. Exact forms of PDF of \(\varphi(I)\) are derived for \(\alpha\) being Cauchy, uniform or Gaussian distribution. These results are based on the extension of Markov-Krein identity and integral representation of a confluent form of the fourth Lauricella function.